KR101610029B1 - Method for a motion estimation based on a variable size block matching and video encoding apparatus using the same - Google Patents

Abstract

The present invention relates to a method and apparatus for reducing the amount of computation in motion estimation based on variable-size block matching for eliminating temporal redundancy of a video image.

A variable-size block matching-based motion estimation method includes: dividing an input frame into partitions according to a plurality of main modes; selecting an optimal one of the plurality of main modes; The main mode or the sub mode of the neighboring partitions existing in the vicinity of the current partition divided according to the selected mode and having a plurality of sub modes, And performing a motion estimation on the current partition in a mode other than the skipped mode among the plurality of submodes.

Variable size block matching, interframe coding, motion estimation, H.264, skip mode

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a motion estimation method based on variable size block matching and a video encoding apparatus using the same,

The present invention relates to a video compression method, and more particularly, to a method and apparatus for reducing computation amount in motion estimation based on variable-size block matching for eliminating temporal redundancy of a video image. will be.

As information and communication technologies including the Internet are developed, not only text and voice but also video communication are increasing. Conventional character - oriented communication methods are not sufficient to satisfy various needs of consumers. Accordingly, multimedia services capable of accommodating various types of information such as text, images, and music are increasing. The amount of multimedia data is so large that it needs a large capacity storage medium and requires a wide bandwidth in transmission. Therefore, it is necessary to use a compression coding technique to transmit multimedia data including characters, images, and audio.

The basic principle of compressing data is the process of eliminating the redundancy of data. A temporal redundancy such as a spatial redundancy such that the same color or object is repeated in an image, a temporal redundancy such as a case where the adjacent frame is almost unchanged in the moving picture frame or the same sound continuously repeats in the audio, The data can be compressed by eliminating the psychological visual overlap considering the insensitivity to the data.

Various video coding standards such as Moving Picture Experts Group (MPEG) -2, MPEG-4, and H.264 have appeared in order to standardize the moving image compression technique. As shown in FIG. 1, all video coding techniques employ a technique called block motion estimation to eliminate temporal redundancy between adjacent video frames. For example, in order to encode a block 30 in the current frame 10, a block 40 matching the block 30 in a reference frame 20 at a different temporal location from the current frame 10 Find. Thereafter, the residual between the block 30 of the current frame 10 and the block 40 of the reference frame 20 is obtained, and the difference is encoded to improve the coding efficiency. Here, the displacement between the blocks is represented as a motion vector, and motion compensation is performed on the reference frame 20 by the motion vector.

H.264, the latest video coding standard, is known to outperform previous standards in terms of coding efficiency. H.264 employs a variable block size ranging from 4x4 to 16x16 for interframe coding, which represents a significant improvement in coding efficiency over conventional techniques for coding fixed sized macroblocks (MB).

In addition, various new technologies such as multiple reference technique, motion vector estimation with 1/4 pixel precision, intra frame coding technique for various directions, and adaptive diblock filtering technique are adopted in H.264.

However, while these new techniques have contributed significantly to improved coding efficiency, there is also a need for much higher computational complexity in H.264 encoders. In particular, the "variable size block matching" technique has been newly introduced in H.264 and contributes significantly to the improvement of coding efficiency, but it causes a relatively large increase in the amount of calculation. The concrete contents of the "variable size block matching" technique will be described below.

In interframe coding, in order to find the size of a block showing optimal coding performance, it is necessary to divide one macroblock into subblocks of various sizes and perform an actual coding for each divided subblock Is required.

The block sizes used in the determination of the interblock mode have five different modes (MD). It includes a mode MD0 indicating "no motion ", a 16x16 block mode MD1, a 16x8 block mode MD2, an 8x16 block mode MD3 and an 8x8 block mode MD8 . Of course, the 8x8 block mode (MD8) can be subdivided into 8x8 (MD4), 8x4 (MD5), 4x8 (MD6) and 4x4 (MD7) block modes. This block mode has a hierarchical structure as shown in FIG. 2 below.

H.264 uses a technique called rate-distortion optimization to determine the optimal block mode among the various sizes of block modes. To perform the rate-distortion optimization for all possible modes, H.264 calculates the coding cost based on the rate-distortion (hereinafter referred to as the rate-distortion cost) associated with each mode, Find the mode that will be the minimum. However, this method can obtain optimum results in terms of visual image quality and coding bit rate, but inevitably requires a large amount of computation.

Thus, in this variable-size block matching technique, the amount of computation can be significantly reduced if it can be skipped without calculating the rate-distortion cost for some block modes. Of course, such a reduction in the calculation amount should be based on the premise that the image quality will not be reduced to a certain level or less. This is because the application of such a variable-size block matching technique itself may be meaningless if the image quality is reduced. As a result, it is necessary to find a method capable of reducing the amount of computation while minimizing the decrease in image quality in motion estimation based on variable size block matching.

It is an object of the present invention to provide a motion estimation technique capable of accurate motion estimation while minimizing the amount of computation.

The technical objects of the present invention are not limited to the technical matters mentioned above, and other technical subjects not mentioned can be clearly understood by those skilled in the art from the following description.

According to an aspect of the present invention, there is provided a variable-size block matching-based motion estimation method including the steps of: (a) dividing an input frame into partitions according to a plurality of main modes; (b) selecting an optimal one of the plurality of main modes; (c) if the selected mode has a plurality of submodes, referring to a type of a main mode or a submode of a neighboring partition existing in the vicinity of a current partition divided according to the selected mode, Skipping at least some of the sub-modes of the sub-mode; And (d) performing motion estimation on the current partition in a mode other than the skipped mode among the plurality of submodes.

According to an aspect of the present invention, there is provided a video encoding apparatus including a partitioning unit dividing an input frame into partitions corresponding to a plurality of main modes; An optimum selection unit for selecting one of the plurality of main modes as an optimum mode; Wherein when the selected mode has a plurality of submodes, the main mode or the submode of the peripheral part existing in the vicinity of the current partition divided according to the selected mode, A skip determining unit skipping at least some of the modes; A motion estimator for performing motion estimation on the current partition in a mode other than the skipped mode among the plurality of submodes; A difference branch for obtaining a residual frame from the input frame by subtracting a motion compensation frame compensated for a reference frame by a motion vector obtained by the motion estimation; And means for coding the residual frame.

According to the present invention, there is an advantage that the amount of computation can be significantly reduced as compared with a conventional motion estimation technique or a conventional variable-size block matching-based motion estimation technique. Nevertheless, since the error of the motion estimation does not occur to a large extent, it is possible to prevent deterioration of image quality due to motion estimation based on the variable-size block matching.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS The advantages and features of the present invention, and the manner of achieving them, will be apparent from and elucidated with reference to the embodiments described hereinafter in conjunction with the accompanying drawings. The present invention may, however, be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art. Is provided to fully convey the scope of the invention to those skilled in the art, and the invention is only defined by the scope of the claims. Like reference numerals refer to like elements throughout the specification.

Referring again to FIG. 2, in the motion estimation of H.264, one macroblock having a size of 16x16 pixels first selects one of MD0, MD1, MD2, MD3, and MD8 based on the rate- (Selection of the first layer). The MD0 means that the motion vector is 0, that is, there is no motion. If a mode other than MD8 is selected, selection of an additional block mode (proceeding to the next layer) is not performed. However, when MD8 is selected, one macroblock is divided into four 8x8 blocks, and one of MD4, MD5, MD6 and MD7 is selected again for each 8x8 block (the second layer Selection). As such, H.264 uses a variable-size block matching of a two-layer structure.

The ratio (%) in which each mode in the first layer is selected for various sample images is as summarized in Table 1 below. In Table 1, the quantization parameter (QP) was selected as 28 for 100 frames of the CIF image.

Table 1 shows that the frequency of MD0 is the lowest and the frequencies of MD1 and MD8 are the highest. In the video coding prior to H.264, motion vectors of 16x16 or 8x8 blocks are searched without selecting a smaller unit. The smaller the block unit, the smaller the error. However, the amount of calculation increases and the number of motion vectors As it would require an appropriate trade-off. In fact, as a whole, the mode that divides into less than MD8 will not be that many, but anyway, in video standards such as H.264, which focuses on the coding efficiency rather than the computational point of view,

Actually, the ratios of the modes belonging to the layer below the MD8 (second layer) are as shown in Tables 2 and 3 (here, QP is selected as 28 for 100 frames of the CIF image). Table 2 shows the cases where MD8 is finally selected optimally, and Table 3 shows cases where MD8 is not optimally selected.

 Looking at Tables 2 and 3, it can be seen that in all cases MD4 occupies the most overwhelming proportion, MD5 and MD6 have similar distribution, and MD7 is least. Therefore, the mode with the smallest influence by skipping will be MD7, followed by MD5 and MD6. Therefore, MD7 must first be considered as a skip object, and then MD5 and MD6 can also be considered. If the skip object is limited to only two modes in order to ensure the accuracy of motion estimation, it is desirable to include MD7 and MD6. In Table 2 and 3, MD5 is slightly higher than MD6, but for many common images, it is preferable to include MD6 as a skipping object rather than MD5 because the horizontal relevance of the pixels is higher than the longitudinal relevance. Particularly, the difference is apparent in the case where the camera moves in the horizontal direction, the object moves in the horizontal direction, and the like. Hereinafter, in one embodiment of the present invention, description will be made assuming that MD6 and MD7 are skipped when a specific condition is satisfied.

3 is a block diagram showing the configuration of a video encoding apparatus 100 that implements a motion estimation method according to an embodiment of the present invention.

The video encoding apparatus 100 includes a partition division unit 110, an optimal selection unit 130, a skip determination unit 140, a motion estimation unit 150, a motion compensation unit 160, a carousel 165, Unit 170, a quantization unit 180, and an entropy encoding unit 190.

The partition division unit 110 divides an input frame into partitions according to a plurality of main modes. The main mode refers to a partition mode belonging to the first layer in a variable size block matching in a hierarchical manner. For example, in FIG. 2, the main modes are MD1, MD2, MD3, and MD8 (except for MD0 without a motion vector). On the other hand, in the case of some of the main modes (in the case of the current H.264, MD8 is applicable), the sub-mode is divided according to the additional sub-mode, and the sub-mode belongs to the second layer. 2, the submodes are MD4, MD5, MD6 and MD7. The distinction here is between MD8 and MD4. MD8 means that it is best to divide into 8x8 four partitions in the main mode once, and each partition is divided into submodes again. As a result, MD4 means only a mode in which the final result according to the submode is 8x8, i.e., does not subdivide.

The optimal selection unit 130 selects one optimal mode among the plurality of main modes and provides the result to the skip determiner 140. In the selection of the mode, the optimum selection unit 130 may be based on a rate-distortion cost calculated by, for example, a linear combination of the error and the bit amount of the image. This rate-distortion cost C can be defined as: < EMI ID = 1.0 >

Here, E represents the error of the image and B represents the bit amount. Specifically, E is a sum of absolute difference (sum of errors between an original image and an image restored after coding) when an interframe is performed based on a block of a specific mode, and B is a sum of absolute difference Lt; / RTI > The parameter λ is a parameter for adjusting whether to emphasize the reduction of the magnitude of the error or the decrease of the magnitude of the bit amount as the Lagrangian coefficient.

For the calculation as shown in Equation (1), the optimum selection unit 130 calculates the bit amount B of the signal output from the quantization unit 180, that is, the encoded data, restores the encoded data, And the error E can be calculated.

As a result, the optimum selection unit 130 selects one mode in which the rate-distortion cost is minimized among the plurality of main modes.

The skip determining unit 140 determines whether or not the neighboring partitions existing in the vicinity of the current partition divided according to the selected mode and already processed in the case where the mode selected by the optimum selecting unit 130 has a plurality of submodes (Hereinafter referred to as "submode skipping" in the present invention) with reference to the main mode or submode type having the plurality of submodes. For example, in the case of H.264, MD8 has multiple submodes, but MD1, MD2 and MD3 are themselves complete and have no additional submodes.

The manner in which the skip determining unit 140 refers to the type of the main mode or the submode of the neighboring partition that has already been processed is basically the case where the neighboring partition is not further divided after division according to the main modes The skip may be performed. In the example of FIG. 2, when the 8x8 partition is used as a reference, when the peripheral partition has any one of MD1, MD2, MD3, and MD4, that is, it means. In FIG. 2, although MD4 is indicated to belong to the second layer, in reality, no further division is made. In order to determine whether the submode is skipped in the current partition of 8x8 size, the referenced peripheral partitions should of course be 8x8 size. However, in practice, a peripheral partition may be larger than an 8x8 size depending on its mode, but it is simpler to think that a larger partition is divided into 8x8 size. For example, in the case of MD1, four 8x8 blocks in a macroblock may be considered to have MD1.

Therefore, the mode that these peripheral partitions can have is one of MD1, MD2, MD3, MD4, MD5, MD6, MD7 (except for MD0). MD8 can be considered to be excluded because it is not the final mode that an 8x8 partition has, and additional submode must be determined. A specific embodiment for determining whether to skip the submode of the current partition from the neighbor partition will be described later with reference to FIGS. 4 and 5. FIG.

The motion estimator 150 performs motion estimation on the current partition in a mode other than the skipped mode (non-skipped submode) among the plurality of submodes. This motion estimation process can determine a motion vector as a block matching algorithm, that is, a displacement in a case where a given motion block is moved on a pixel-by-pixel basis within a specific search area of a reference frame and the error is lowest. As a criterion for measuring the error, SAD (sum of absolute differences) is generally used. As a result, the optimum mode among the modes including the main mode and the non-skipped submode is finally selected, and the selected final mode and the motion vector at that time are the final outputs of the motion estimation unit 150. [

The motion compensating unit 160 obtains a motion compensated frame by performing motion compensation on the reference frame using the selected final mode and the motion vector provided from the motion estimating unit 150. [

On the other hand, in the input frame, the difference branch 165 obtains a residual frame by subtracting the motion compensation frame compensated for the reference frame by the motion vector. In an embodiment of the present invention, the spatial transform unit 170, the quantization unit 180, and the entropy coding unit 190 are used as means for coding the residual frame.

The spatial transform unit 170 transforms the residual frame into a frequency domain using a predetermined spatial transform method. These spatial transformations are mainly DCT (Discrete Cosine Transform), and sometimes wavelet transforms are used. The coefficients obtained as a result of spatial conversion are called conversion coefficients. When DCT is used as a spatial transform, DCT coefficients are used. When wavelet transform is used, wavelet coefficients are used.

The quantization unit 180 quantizes the transform coefficients obtained by the spatial transform unit 170. The term 'quantization' refers to a process of dividing the transformation coefficient expressed by an arbitrary real number value into a predetermined section, representing the transformation coefficient as a discrete value, and matching it with a predetermined index. In particular, when wavelet transform is used as a spatial transform method, embedded quantization is also used as a quantization method.

The entropy encoding unit 190 losslessly encodes the transform coefficients quantized by the quantization unit 180 and the final mode and the motion vector provided by the motion estimation unit 150 to generate an output bitstream. Such lossless coding methods include arithmetic coding, variable length coding, Huffman coding, and the like.

Each of the components of FIG. 3 may denote software or hardware such as a field-programmable gate array (FPGA) or an application-specific integrated circuit (ASIC). However, the components are not limited to software or hardware, and may be configured to be in an addressable storage medium and configured to execute one or more processors. The functions provided in the components may be implemented by a more detailed component or may be implemented by a single component that performs a specific function by combining a plurality of components.

4 and 5 are diagrams for explaining a specific algorithm for implementing "submode skipping" according to an embodiment of the present invention.

If the current MB is determined to be MD1, MD2 or MD3 instead of MD8, then it is not necessary to consider the submode itself, so it is not necessary to consider the submode skipping proposed in the present invention. However, if the current MB is MD8, the current MB is divided into four 8x8 blocks, and the sub mode skip is determined for each 8x8 block.

Referring to FIG. 4, there are four types of neighboring partitions A, B, C, and D that are referred to in order to determine whether a current macroblock MB skips a submode. Case A is the case where the current macro block is tangent to the left boundary of the frame or slice, Case B is the case where the current macroblock is tangent to the upper boundary of the frame or slice, Case C is the case where the current macroblock is located at the right boundary of the frame or slice And the like. In general, when the current macroblock is not in contact with the boundary, all four neighboring macroblocks can be referred to as the type D. Note that even if all of the neighboring macroblocks shown in Fig. 4 are divided into four partitions, not all of them are divided into MD8, and a number for identifying a partition to be referenced is added, I do not know. As described above, assuming that the upper left macroblock is MD1, all four partitions that may be considered to be included therein are all considered to be MD1.

However, if it is assumed that the neighboring macroblock and the current macroblock are divided into four 8x8 partitions, they can be sequentially numbered from 0 to 3.

5 is a diagram illustrating directions in which four partitions belonging to a current macroblock refer to neighboring partitions. The basic concepts of such references are shown in (a) and (b). (a) shows a direction referencing from a neighboring partition, and (b) shows a direction referring to another partition in the current macroblock.

(c) is an example of a case of referring in the vertical direction in (a) and (b). (d) is an example of a case of referring in the horizontal direction in (a) and (b). (e) is an example of a case where reference is made in the rightward and downward diagonal directions in (a) and (b). (F) is an example of a case where reference is made in the left-down diagonal direction in (a) and (b).

For example, when referring to the numbering in FIG. 4, the neighboring partitions that can be referred to by the third partition of the current macroblock are as follows. According to (c), it is possible to refer to the partition 3 of the upper macroblock and the partition 1 of the current macroblock. According to (d), the partition 3 of the left macro block and the partition 2 Can be referred to. According to (e), it is possible to refer to the partition # 3 of the upper left side macroblock and the partition # 0 of the current macroblock, and according to (f), the partition # 3 of the upper right side macroblock can be referred to. Therefore, there are seven referenceable partitions for applying the submode skip to the current macroblock. However, it is needless to say that all such referenceable partitions are not actually referred to, and all or some of them may be referred to as necessary. The remaining 0, 1, and 2 partitions of the current macroblock can also refer to neighboring partitions to apply the submode skip.

6 and 7 below are pseudo codes for applying the submode skip. Here, the meanings of terms used are shown in Table 4 below.

Terms meaning CMB Current macroblock LMB Based on the current macroblock, C8B Number of the 8x8 partition of the current macroblock LRM Left reference mode UMB Based on the current macroblock, the upper macroblock URM Upper reference mode LUMB Based on the current macroblock, the upper left side macroblock LURM Upper left reference mode RUMB Based on the current macroblock, the upper right macroblock RURM Right-side reference mode C8B0 Optimal mode of an 8x8 partition at the 0 position of the current macroblock C8B1 Optimum mode of 8x8 partition at 1 of current macroblock C8B2 Optimal mode of 8x8 partition at position 2 of the current macroblock

First, FIG. 6 shows an example of a pseudo code in which the reference method according to FIGS. 4 and 5 is implemented as a code. Referring to FIG. 6, in consideration of the boundary conditions of FIG. 4, it can be seen that the neighboring partitions store modes (optimally determined modes) for reference in the same manner as FIG.

FIG. 7 illustrates an example of a pseudo code for determining whether to skip MD6 and MD7 in a target 8x8 partition (current partition) using the mode of the stored neighboring partitions as shown in FIG. In summary, it can be seen that submode skipping (skipping only MD6 and MD7 in the submode in this example) applies to the current partition only if all partitions referenced by the current partition have MD1 or MD4. Of course, since this is an embodiment only, the types of submodes to be skipped may be further added or reduced. In the case of having MD1, MD2, MD3 and MD4 as well as MD1 or MD4, .

In the last two lines of FIG. 7, the marked portion means that at least MD7 will skip if a certain condition is satisfied, even if the submode skip is determined not to be applied. The judgment condition is met if the rate-distortion cost (RDC) of MD5 is greater than the rate-distortion cost (RDC) of MD4. When these conditions are satisfied, it means that MD4 is more optimal than MD5. Therefore, in this case, since it means that there is a high possibility that the 8x8 partition is determined more optimally than the partition where the partition is divided, the possibility of optimal determination in MD7 is very rare. Therefore, such a skip of MD7 is possible.

Table 5 below shows changes in picture quality (PSNR), bit rate (BR) and coding time (T) as a result of applying the submode skip according to the present invention to various sequences. For these sequences, 100 frames of CIF image were used, and the quantization parameter was selected as 28. In addition, the number of frames to be referred to is 5, and a fast full search is applied.

Referring to Table 5, when the present invention is applied, on average, the PSNR is reduced by 0.001% and the bit rate is increased by 0.22%, while the coding time can be reduced by 4.5%. This means that not only the motion estimation process but also the time saved in the entire video coding, the coding time reduction ratio is much larger if only the motion estimation process is compared.

While the present invention has been described in connection with what is presently considered to be practical exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, You will understand. It is therefore to be understood that the embodiments described above are in all respects illustrative and not restrictive.

1 is a diagram showing a basic concept of interframe coding using a block motion estimation technique.

FIG. 2 is a diagram showing a plurality of block modes according to a hierarchical division method defined in H.264.

3 is a block diagram illustrating a configuration of a video encoding apparatus for implementing a motion estimation method according to an embodiment of the present invention.

FIG. 4 is a diagram showing that the number of neighboring macroblocks to be referred to can be varied according to the position of the current macroblock.

5 is a diagram illustrating directions in which four partitions belonging to a current macroblock refer to neighboring partitions.

FIG. 6 is a diagram showing an example of a pseudo code that implements the reference method according to FIGS. 4 and 5 as a code.

FIG. 7 is a diagram illustrating an example of a pseudo code for determining whether to skip a submode in a current partition, which is actually a target, using a mode of neighboring partitions stored as shown in FIG.

(Reference Numerals for Main Parts of the Drawings)

100: Video encoding device 110: Partitioning division

130: optimal selection unit 140: skip determination unit

150: Motion estimation unit 160: Motion compensation unit

165: Sub-branch 170: Spatial transformation unit

180: quantization unit 190: entropy coding unit

Claims (19)

(a) partitioning an input frame into partitions according to a plurality of main modes; (b) selecting an optimal one of the plurality of main modes; (c) if the selected mode has a plurality of submodes, referring to a type of a main mode or a submode of a neighboring partition existing in the vicinity of the current partition divided according to the selected mode, Skipping the selected mode if a skipable mode is not found among a plurality of submodes, and skipping the selected mode if a skipable mode is found among the plurality of submodes; And (d) performing motion estimation on the current partition in a mode other than the skipped mode among the plurality of submodes, Whether the mode is the skip mode, Wherein the motion vector is set with reference to the frequency of occurrence of the plurality of submodes in advance. 2. The method of claim 1, wherein step (b) And selecting the optimal one of the modes based on a rate-distortion cost calculated as a linear combination of error and bit amount of the image. The method of claim 1, wherein the plurality of main modes 16x16 partition mode (MD1), 16x8 partition mode (MD2), 8x16 partition mode (MD3), and 8x8 partition mode (MD8) And when the selected mode has a plurality of submodes, the selected mode is MD8. 2. The method of claim 1, wherein the plurality of submodes A method of motion estimation based on variable size block matching, including an 8x8 partition mode (MD4), an 8x4 partition mode (MD5), a 4x8 partition mode (MD6), and a 4x4 partition mode (MD7). 5. The method of claim 4, wherein the skip mode is at least And the MD7. 6. The method of claim 5, wherein the skip mode is And further comprising the MD6. The method of claim 1, Wherein the at least one of the left partition, the upper partition, the upper left partition, and the upper right partition is located locally with respect to the current partition. 2. The method of claim 1, wherein step (c) Wherein the motion compensation is performed only when the neighboring partitions are not further divided after being divided according to the main modes. 9. The method of claim 8, wherein step (c) Wherein the peripheral partition has a 16x16 partition mode (MD1) and an 8x8 partition mode (MD4). A partitioning unit for dividing an input frame into partitions according to a plurality of main modes; An optimum selection unit for selecting one of the plurality of main modes as an optimum mode; The main mode or the sub mode of the peripheral part existing in the vicinity of the current partition divided according to the selected mode in the case that the selected mode has a plurality of sub modes, A skip determining unit skipping the selected mode if a skipable mode is not found in the mode, and skipping the selected mode if a skipable mode is found among the plurality of submodes; A motion estimator for performing motion estimation on the current partition in a mode other than the skipped mode among the plurality of submodes; A difference branch for obtaining a residual frame from the input frame by subtracting a motion compensation frame compensated for a reference frame by a motion vector obtained by the motion estimation; And And means for coding the residual frame, Whether the mode is the skip mode, Is set with reference to the frequency of occurrence of the plurality of submodes in advance. 11. The apparatus of claim 10, wherein the means for coding the residual frame comprises: A spatial transformer for transforming the residual frame into a frequency domain and generating a transform coefficient; A quantization unit for quantizing the transform coefficient; And And an entropy encoding unit for lossless encoding the result of the quantization and the motion vector. 11. The apparatus of claim 10, wherein the optimal selection unit And selects the optimal one mode based on a rate-distortion cost computed as a linear combination of error and bit amount of the image. 11. The method of claim 10, wherein the plurality of main modes 16x16 partition mode (MD1), 16x8 partition mode (MD2), 8x16 partition mode (MD3), and 8x8 partition mode (MD8) And when the selected mode has a plurality of submodes, it means that the selected mode is MD8. 11. The method of claim 10, wherein the plurality of submodes A video encoding device including an 8x8 partition mode (MD4), an 8x4 partition mode (MD5), a 4x8 partition mode (MD6), and a 4x4 partition mode (MD7). 15. The method of claim 14, wherein the skip mode is at least And the MD7. 16. The method of claim 15, wherein the skip mode is Further comprising the MD6. 11. The system of claim 10, Wherein the at least one of the left partition, the upper partition, the upper left partition, and the upper right partition is located locally with respect to the current partition. 11. The apparatus of claim 10, wherein the skip determiner And performs the skip only when the peripheral partition is not further divided after division according to the main modes. 19. The apparatus of claim 18, wherein the skip determining unit Wherein the skipping is performed only when the peripheral partition has a 16x16 partition mode (MD1) and an 8x8 partition mode (MD8).

Images